diff --git a/official/nlp/ernie/README_CN.md b/official/nlp/ernie/README_CN.md
index a21a440fb357348feb62aa9b85082f79543e64b2..1abbe73f11c3202b0dc5506a5b5bbd61f137d0f2 100644
--- a/official/nlp/ernie/README_CN.md
+++ b/official/nlp/ernie/README_CN.md
@@ -10,11 +10,11 @@
 - [环境要求](#环境要求)
 - [快速入门](#快速入门)
     - [脚本说明](#脚本说明)
-    - [脚本和样例代码](#脚本和样例代码)
+    - [脚本和代码](#脚本和代码)
     - [选项及参数](#选项及参数)
         - [选项](#选项)
         - [参数](#参数)
-    - [训练过程](#训练过程)
+    - [预训练过程](#预训练过程)
         - [用法](#用法)
             - [下载数据集并预处理](#下载数据集并预处理)
             - [Ascend处理器上运行](#ascend处理器上运行)
@@ -24,13 +24,26 @@
             - [Ascend处理器上运行单卡微调](#ascend处理器上运行单卡微调)
             - [Ascend处理器上单机多卡微调](#ascend处理器上单机多卡微调)
             - [Ascend处理器上运行微调后的模型评估](#ascend处理器上运行微调后的模型评估)
+    - [导出onnx模型](#导出onnx模型)
+        - [onnx模型导出](#onnx模型导出)
+        - [onnx模型评估](#onnx模型评估)
+            - [基于chnsenticorp数据集进行onnx评估](#基于chnsenticorp数据集进行onnx评估)
+            - [基于xnli数据集进行onnx评估](#基于xnli数据集进行onnx评估)
+            - [基于dbqa数据集进行onnx评估](#基于dbqa数据集进行onnx评估)
     - [导出mindir模型](#导出mindir模型)
     - [推理过程](#推理过程)
         - [用法](#用法-2)
         - [结果](#结果)
-    - [精度与性能](#精度与性能)
-        - [推理性能](#推理性能)
+    - [模型描述](#模型描述)
+        - [精度与性能](#精度与性能)
+            - [推理性能](#推理性能)
+                - [命名实体识别任务](#命名实体识别任务)
+                - [情感分析任务](#情感分析任务)
+                - [自然语言接口](#自然语言接口)
+                - [问答](#问答)
+                - [阅读理解](#阅读理解)
 - [ModelZoo主页](#modelzoo主页)
+- [FAQ](#faq)
 
 <!-- /TOC -->
 
@@ -77,9 +90,9 @@ bash scripts/download_datasets.sh pretrain
 
 # 将数据集转为MindRecord
 # 预训练数据集
-bash scripts/convert_pretrain_dataset.sh /path/zh_wiki/ /path/zh_wiki/mindrecord/
+bash scripts/convert_pretrain_datasets.sh /path/zh_wiki/ /path/zh_wiki/mindrecord/
 # 微调数据集
-bash scripts/convert_finetune_dataset.sh /path/msra_ner/ /path/msra_ner/mindrecord/ msra_ner
+bash scripts/convert_finetune_datasets.sh /path/msra_ner/ /path/msra_ner/mindrecord/ msra_ner
 
 # 单机运行预训练示例
 bash scripts/run_standalone_pretrain_ascend.sh 0 1 /path/cn-wiki-128
@@ -112,6 +125,7 @@ bash scripts/run_distribute_finetune.sh rank_table.json xnli
     ├─export.sh                               # 导出模型中间表示脚本，如MindIR
     ├─migrate_pretrained_models.sh            # 在x86设备上将Paddle预训练权重参数转为MindSpore权重参数脚本
     ├─run_distribute_finetune.sh              # Ascend设备上多卡运行微调任务脚本
+    ├─run_eval_onnx.sh                        # 对导出的onnx模型评估脚本
     ├─run_finetune_eval.sh                    # Ascend设备上测试微调结果脚本
     ├─run_infer_310.sh                        # Ascend 310设备推理脚本
     ├─run_standalone_finetune.sh              # Ascend设备上单卡运行微调任务脚本
@@ -138,7 +152,8 @@ bash scripts/run_distribute_finetune.sh rank_table.json xnli
   ├─run_ernie_classifier.py                   # 分类器任务的微调和评估网络
   ├─run_ernie_mrc.py                          # 阅读理解任务的微调和评估网络
   ├─run_ernie_ner.py                          # NER任务的微调和评估网络
-  └─run_ernie_pretrain.py                     # 预训练网络
+  ├─run_ernie_pretrain.py                     # 预训练网络
+  └─run_eval_onnx.py                          # 评估onnx模型
 ```
 
 ## 选项及参数
@@ -215,15 +230,19 @@ bash scripts/download_datasets.sh pretrain
 然后进行数据预处理，对文本进行分词，并随机mask词语：
 
 ```bash
-bash scripts/convert_pretrain_dataset.sh /path/zh_wiki/ /path/zh_wiki/mindrecord/
+bash scripts/convert_pretrain_datasets.sh /path/zh_wiki/ /path/zh_wiki/mindrecord/
 ```
 
 > **注意：**
+>
 > 1. 维基百科文本抽取依赖`wikiextractor`，数据预处理依赖结巴分词和`OpenCC`繁简体转换，可以通过以下命令安装依赖：
->```bash
->   pip install -r requirements.txt
->```
+>
+> ```bash
+> pip install -r requirements.txt
+> ```
+>
 > 2. 若需要使用私有词典进行分词，可修改`scr/pretrain_reader.py`中`get_word_segs`方法。
+
 #### Ascend处理器上运行
 
 ```bash
@@ -284,6 +303,7 @@ bash scripts/run_distribute_finetune.sh [RANK_TABLE_FILE] [TASK_TYPE]
 以上命令后台运行，您可以在{task_type}_train_log.txt中查看训练日志。
 
 > **注意：**
+>
 > 1. `rank_table.json`可以通过`/etc/hccn.conf`获取加速卡IP进行配置。
 > 2. `drcd, cmrc`数据集评估需要使用`nltk`，请通过以下命令安装并下载依赖库，然后运行微调脚本。
 
@@ -324,6 +344,93 @@ F1 0.920507
 {"exact_match": 84.13970798740338, "f1": 90.52935807300771}
 ```
 
+## 导出onnx模型
+
+### finetune数据集转化为mindspore格式
+
+```shell
+bash scripts/convert_finetune_datasets.sh [DATASET_PATH] [OUTPUT_PATH] [TASK_TYPE]
+# bash scripts/convert_finetune_datasets.sh data/chnsenticorp/ data/chnsenticorp/ chnsenticorp
+# bash scripts/convert_finetune_datasets.sh data/xnli/ data/xnli/ xnli
+# bash scripts/convert_finetune_datasets.sh data/nlpcc-dbqa/ data/nlpcc-dbqa/ dbqa
+```
+
+[DATASET_PATH]：数据集路径； [OUTPUT_PATH]：转换后输出路径； [TASK_TYPE]：任务类型
+
+### onnx模型导出
+
+```bash
+python export.py --ckpt_file [CKPT_PATH] --file_format onnx --file_name [FILE_NAME] --device_target GPU --task_type [TASK_TYPE] --number_labels [NUMBER_LABELS]
+# chnsenticorp数据集
+# python export.py --ckpt_file ./save_models/chnsenticorp-0-3_300.ckpt --file_format ONNX --file_name ernie_finetune --device_target GPU --task_type chnsenticorp --number_labels 3
+
+# xnli数据集
+# python export.py --ckpt_file ./save_models/ernie_ascend_v170_xnli_official_nlp_acc77.94.ckpt --file_format ONNX --device_target GPU --task_type xnli --number_labels 3
+
+# dbqa数据集
+# python export.py --ckpt_file ./save_models/ernie_ascend_v170_bdqa_official_nlp_F1score84.96.ckpt --file_format ONNX --device_target GPU --task_type dbqa --number_labels 2
+```
+
+[CKPT_PATH]：为ckpt路径；
+
+[FILE_NAME]：为导出onnx文件名称；
+
+[TASK_TYPE]：任务数据集名称，包括`[chnsenticorp, xnli, dbqa]`
+
+[NUMBER_LABELS]：分类任务label数量， chnsenticorp、xnli为3， dbqa为2
+
+### onnx模型评估
+
+```bash
+bash scripts/run_eval_onnx.sh [TASK_TYPE]
+# bash scripts/run_eval_onnx.sh chnsenticorp
+# TASK_TYPE including [chnsenticorp, xnli, dbqa]
+```
+
+以上命令后台运行，您可以在`./ms_log/[TASK_TYPE]_onnx_log.txt`查看运行结果。
+
+onnx模型路径为 `ONNX_PATH=${CUR_DIR}/ernie_finetune.onnx`
+
+eval数据集路径为 `EVAL_DATA_PATH=${CUR_DIR}/data/${TASK_TYPE}/${TASK_TYPE}_test.mindrecord`
+    如：`home/models/official/nlp/ernie/data/chnsenticorp/chnsenticorp_test.mindrecord`
+
+具体请看 scripts/run_eval_onnx.sh 脚本
+
+#### 基于chnsenticorp数据集进行onnx评估
+
+```bash
+bash scripts/run_eval_onnx.sh chnsenticorp
+
+cat ms_log/chnsenticorp_onnx_log.txt
+==============================================================
+acc_num 1150 , total_num 1200, accuracy 0.958333
+==============================================================
+```
+
+#### 基于xnli数据集进行onnx评估
+
+```bash
+bash scripts/run_eval_onnx.sh xnli
+
+cat ms_log/xnli_onnx_log.txt
+==============================================================
+acc_num 3902 , total_num 5010, accuracy 0.778842
+==============================================================
+```
+
+#### 基于dbqa数据集进行onnx评估
+
+```bash
+bash scripts/run_eval_onnx.sh dbqa
+
+cat ms_log/dbqa_onnx_log.txt
+==============================================================
+Precision 0.804701
+Recall 0.897794
+F1 0.848703
+==============================================================
+```
+
 ## 导出mindir模型
 
 ```bash
diff --git a/official/nlp/ernie/requirements.txt b/official/nlp/ernie/requirements.txt
index 6e682e5e1249910f762ebd4e521b6ddd0d80234c..095d2da4907c3c1c0d147806165bfa39ea75d734 100644
--- a/official/nlp/ernie/requirements.txt
+++ b/official/nlp/ernie/requirements.txt
@@ -2,4 +2,5 @@ paddlepaddle
 nltk
 wikiextractor
 jieba
-opencc-python-reimplemented
\ No newline at end of file
+opencc-python-reimplemented
+onnxruntime-gpu
\ No newline at end of file
diff --git a/official/nlp/ernie/run_eval_onnx.py b/official/nlp/ernie/run_eval_onnx.py
new file mode 100644
index 0000000000000000000000000000000000000000..f33eb9db127ae58d9b0a8ef912175297153481dc
--- /dev/null
+++ b/official/nlp/ernie/run_eval_onnx.py
@@ -0,0 +1,133 @@
+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+'''
+ERNIE task classifier onnx script.
+'''
+
+import os
+import argparse
+from src.dataset import create_finetune_dataset
+from src.assessment_method import Accuracy, F1
+from mindspore import Tensor, dtype
+import onnxruntime as ort
+
+
+def eval_result_print(assessment_method="accuracy", callback=None):
+    """ print eval result """
+    if assessment_method == "accuracy":
+        print("acc_num {} , total_num {}, accuracy {:.6f}".format(callback.acc_num, callback.total_num,
+                                                                  callback.acc_num / callback.total_num))
+    elif assessment_method == "f1":
+        print("Precision {:.6f} ".format(callback.TP / (callback.TP + callback.FP)))
+        print("Recall {:.6f} ".format(callback.TP / (callback.TP + callback.FN)))
+        print("F1 {:.6f} ".format(2 * callback.TP / (2 * callback.TP + callback.FP + callback.FN)))
+    else:
+        raise ValueError("Assessment method not supported, support: [accuracy, f1]")
+
+def create_session(checkpoint_path, target_device):
+    """Create ONNX runtime session"""
+    if target_device == 'GPU':
+        providers = ['CUDAExecutionProvider']
+    elif target_device == 'CPU':
+        providers = ['CPUExecutionProvider']
+    else:
+        raise ValueError(
+            f'Unsupported target device {target_device}, '
+            f'Expected one of: "CPU", "GPU"'
+        )
+    session = ort.InferenceSession(checkpoint_path, providers=providers)
+    input_name_0 = session.get_inputs()[0].name
+    input_name_1 = session.get_inputs()[1].name
+    input_name_2 = session.get_inputs()[2].name
+    output_name_0 = session.get_outputs()[0].name
+    return session, input_name_0, input_name_1, input_name_2, output_name_0
+
+
+def do_eval_onnx(dataset=None, onnx_file_name="", num_class=3, assessment_method="accuracy", target_device="GPU"):
+    """ do eval for onnx model"""
+    if assessment_method == "accuracy":
+        callback = Accuracy()
+    elif assessment_method == "f1":
+        callback = F1(num_class)
+    else:
+        raise ValueError("Assessment method not supported, support: [accuracy, f1]")
+
+    columns_list = ["input_ids", "input_mask", "token_type_id", "label_ids"]
+
+    if not os.path.exists(onnx_file_name):
+        raise ValueError("ONNX file not exists, please check onnx file has been saved and whether the "
+                         "export_file_name is correct.")
+
+    session, input_name_0, input_name_1, input_name_2, output_name_0 = create_session(onnx_file_name, target_device)
+
+
+    for data in dataset.create_dict_iterator(num_epochs=1):
+        input_data = []
+        for i in columns_list:
+            input_data.append(data[i])
+        input_ids, input_mask, token_type_id, label_ids = input_data
+
+        x0 = input_ids.asnumpy()
+        x1 = input_mask.asnumpy()
+        x2 = token_type_id.asnumpy()
+
+        result = session.run([output_name_0], {input_name_0: x0, input_name_1: x1, input_name_2: x2})
+        logits = Tensor(result[0], dtype.float32)
+        callback.update(logits, label_ids)
+
+    print("==============================================================")
+    eval_result_print(assessment_method, callback)
+    print("==============================================================")
+
+
+def run_classifier_onnx():
+    """run classifier task for onnx model"""
+    args_opt = parse_args()
+    if args_opt.eval_data_file_path == "":
+        raise ValueError("'eval_data_file_path' must be set when do onnx evaluation task")
+    assessment_method = args_opt.assessment_method.lower()
+    ds = create_finetune_dataset(batch_size=args_opt.eval_batch_size,
+                                 repeat_count=1,
+                                 data_file_path=args_opt.eval_data_file_path,
+                                 do_shuffle=(args_opt.eval_data_shuffle.lower() == "true"))
+    do_eval_onnx(ds, args_opt.onnx_file, args_opt.number_labels, assessment_method, args_opt.device_target)
+
+
+def parse_args():
+    """set and check parameters."""
+    parser = argparse.ArgumentParser(description="run classifier")
+    parser.add_argument("--task_type", type=str, default="chnsenticorp", choices=["chnsenticorp", "xnli", "dbqa"],
+                        help="Task type, default is chnsenticorp")
+    parser.add_argument("--assessment_method", type=str, default="accuracy", choices=["accuracy", "f1"],
+                        help="Assessment method")
+    parser.add_argument("--device_target", type=str, default="GPU", choices=["Ascend", "GPU"],
+                        help="Device type, default is Ascend")
+    parser.add_argument('--onnx_file', type=str,
+                        default="./ernie_finetune.onnx",
+                        help='Onnx file path')
+    parser.add_argument("--device_id", type=int, default=0, help="Device id, default is 0.")
+    parser.add_argument("--number_labels", type=int, default=3, help="The number of class, default is 3.")
+    parser.add_argument("--eval_data_shuffle", type=str, default="false", choices=["true", "false"],
+                        help="Enable eval data shuffle, default is false")
+    parser.add_argument("--eval_batch_size", type=int, default=1, help="Eval batch size, default is 1")
+    parser.add_argument("--eval_data_file_path", type=str, default="",
+                        help="Data path, it is better to use absolute path")
+    args_opt = parser.parse_args()
+
+    return args_opt
+
+
+if __name__ == "__main__":
+    run_classifier_onnx()
diff --git a/official/nlp/ernie/scripts/run_eval_onnx.sh b/official/nlp/ernie/scripts/run_eval_onnx.sh
new file mode 100644
index 0000000000000000000000000000000000000000..5cfdde2ff765662fd344e36eb2a3ff7a7dfaf6a2
--- /dev/null
+++ b/official/nlp/ernie/scripts/run_eval_onnx.sh
@@ -0,0 +1,70 @@
+#!/bin/bash
+# Copyright 2022 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+if [ $# -ne 1 ]
+then
+    echo "=============================================================================================================="
+    echo "Please run the script as: "
+    echo "bash run_eval_onnx.sh [TASK_TYPE]"
+    echo "for example: bash scripts/run_eval_onnx.sh chnsenticorp"
+    echo "TASK_TYPE including [chnsenticorp, xnli, dbqa]"
+    echo "=============================================================================================================="
+exit 1
+fi
+
+TASK_TYPE=$1
+DEVICE_ID=0
+CUR_DIR=`pwd`
+mkdir -p ms_log
+ONNX_PATH=${CUR_DIR}/ernie_finetune.onnx
+DATA_PATH=${CUR_DIR}/data
+GLOG_log_dir=${CUR_DIR}/ms_log
+
+case $TASK_TYPE in
+  "chnsenticorp")
+    PY_NAME=run_eval_onnx
+    NUM_LABELS=3
+    EVAL_BATCH_SIZE=1
+    EVAL_DATA_PATH="${DATA_PATH}/chnsenticorp/chnsenticorp_test.mindrecord"
+    ASSESSMENT_METHOD="accuracy"
+    ;;
+
+  "xnli")
+    PY_NAME=run_eval_onnx
+    NUM_LABELS=3
+    EVAL_BATCH_SIZE=1
+    EVAL_DATA_PATH="${DATA_PATH}/xnli/xnli_test.mindrecord"
+    ASSESSMENT_METHOD="accuracy"
+    ;;
+
+  "dbqa")
+    PY_NAME=run_eval_onnx
+    NUM_LABELS=2
+    EVAL_BATCH_SIZE=1
+    EVAL_DATA_PATH="${DATA_PATH}/nlpcc-dbqa/dbqa_test.mindrecord"
+    ASSESSMENT_METHOD="f1"
+    ;;
+esac
+
+python ${CUR_DIR}/${PY_NAME}.py \
+    --task_type=${TASK_TYPE} \
+    --device_target="GPU" \
+    --device_id=${DEVICE_ID} \
+    --number_labels=${NUM_LABELS} \
+    --eval_data_shuffle="false" \
+    --eval_batch_size=${EVAL_BATCH_SIZE} \
+    --eval_data_file_path=${EVAL_DATA_PATH} \
+    --onnx_file=${ONNX_PATH} \
+    --assessment_method=${ASSESSMENT_METHOD} > ${GLOG_log_dir}/${TASK_TYPE}_onnx_log.txt 2>&1 &
diff --git a/official/nlp/ernie/src/dataset.py b/official/nlp/ernie/src/dataset.py
index d672304a316e6ab75dfc1a10c2e8ab69ef8ec27d..5aa74a06353c84c17dca69f501ba3ddb7569bd04 100644
--- a/official/nlp/ernie/src/dataset.py
+++ b/official/nlp/ernie/src/dataset.py
@@ -55,6 +55,7 @@ def create_finetune_dataset(batch_size=1,
                             do_shuffle=True):
     """create finetune or evaluation dataset"""
     type_cast_op = C.TypeCast(mstype.int32)
+
     data_set = ds.MindDataset(data_file_path,
                               columns_list=["input_ids", "input_mask", "token_type_id", "label_ids"],
                               shuffle=do_shuffle,
@@ -79,6 +80,7 @@ def create_mrc_dataset(batch_size=1,
                        drop_reminder=False):
     """create finetune or evaluation dataset"""
     type_cast_op = C.TypeCast(mstype.int32)
+
     if is_training:
         data_set = ds.MindDataset(data_file_path,
                                   columns_list=["input_ids", "input_mask", "token_type_id",
diff --git a/official/nlp/ernie/src/finetune_task_reader.py b/official/nlp/ernie/src/finetune_task_reader.py
index 868397d7065b6f5effe53d8171cb1c1eb237dc46..fdef43a87f5abdb1d08227097c0269d85f725416 100644
--- a/official/nlp/ernie/src/finetune_task_reader.py
+++ b/official/nlp/ernie/src/finetune_task_reader.py
@@ -23,7 +23,7 @@ from collections import namedtuple
 import numpy as np
 from mindspore.mindrecord import FileWriter
 from mindspore.log import logging
-from src.tokenizer import FullTokenizer, convert_to_unicode, tokenize_chinese_chars
+from .tokenizer import FullTokenizer, convert_to_unicode, tokenize_chinese_chars
 
 def csv_reader(fd, delimiter='\t'):
     """
diff --git a/official/nlp/ernie/src/pretrain_reader.py b/official/nlp/ernie/src/pretrain_reader.py
index b42898f7fa313cef6c93a2402b4456e6002b6d11..e200b66cebde6646068ad971677f3ccc1dbcc96e 100644
--- a/official/nlp/ernie/src/pretrain_reader.py
+++ b/official/nlp/ernie/src/pretrain_reader.py
@@ -24,8 +24,8 @@ import jieba
 import numpy as np
 from opencc import OpenCC
 from mindspore.mindrecord import FileWriter
-from src.tokenizer import convert_to_unicode, CharTokenizer
-from src.utils import get_file_list
+from .tokenizer import convert_to_unicode, CharTokenizer
+from .utils import get_file_list
 
 class ErnieDataReader:
     """Ernie data reader"""